SGLang's MegaKernel+PD Boosts DeepSeek Inference Performance
SGLang unveiled optimizations for DeepSeek V4.1 Flash, raising single-user speed to 943 tok/s and per-GPU throughput to 654K tok/s, while its MegaKernel+PD disaggregation solution with RadixArk improves both latency and throughput with much lower TCO.
2026-10-27 ~ 2026-10-27 · 2 related posts
- SGLang MegaKernel + PD Pushes the TCO-Latency Frontier on GB300, SemiAnalysis Benchmarks Show — AccBalanced · 2026-10-27
- SGLang doubles DeepSeek V4.1 Flash serving: 943 tok/s/user latency, 654K tok/s per GPU — ying11231 · 2026-10-27